Surveying radiologists to determine the value of AI-based automated measurements and reporting of organ volumes in the clinical workflow

Purpose: While extensive work has been done in artificial intelligence (AI) algorithm development, examples of their clinical deployment are fewer. Clinical AI applications must satisfy stakeholder needs, given both excitement and skepticism about their use. We developed a cloud-based platform for automatically analyzing abdominal CT using an AI-based machine learning (ML) approach, reporting results directly to clinical PACS.

Methods/Materials: This research was IRB-approved. We deployed a survey using the Qualtrics online survey platform (Silver Lake LLC, Seattle, WA) with 25 questions divided into 7 sections. 30 radiologists, including residents and fellows, voluntarily participated in this survey. Survey takers received an email with an online survey link and were de-identified. Survey results were analyzed using JMP Pro 16 (SAS Inc., Cary, NC). Contingency analysis was performed to determine associations between two categorical variables using a chi-square test. Question responses were plotted on a Mosaic plot to identify extreme variations in responses to target areas for further development.

Results: Each responder answered the 25 questions in less than 15 minutes on either a computer or mobile device. Contingency analysis depicted that 3 of 5 responders desired additional education regarding how AI results would be used for patient care. 70% of responders were ambivalent or slightly negative about whether race and ethnicity data should be included in automated reports. All responders agreed that AI results would be useful for research and quality improvement. All responders agreed that there should be a mechanism for radiologists to provide feedback on AI-generated results. These results will be used to improve physician education and the AI platform.

Conclusions: Our software platform delivers the results of multiple AI algorithms (liver/spleen volume, visceral/subcutaneous fat, hepatic steatosis determination, and muscle volume), including an image overlay and quantitative measurements, automatically to PACS for radiologist review. We present the results from a radiologist survey to determine the value of AI platform deployment in routine clinical workflow.